(573g) The Landscape of Drug Sensitivity in Cancer Cell Lines Reveals Effective Drug Combinations for Cancer

Targeted therapeutics have revolutionized cancer treatment. However, because only seven percent of cancer patients respond to molecularly targeted therapies, improved methods to match tumors with effective therapeutics are desperately needed to realize the promise of precision oncology. In addition, because acquired resistance is almost universally observed in single-agent treatments, it is crucial to identify drug combinations which may increase the efficacy of targeted therapies and/or prevent the emergence of resistant cells. Here, we demonstrate that a bioinformatic algorithm recently developed by our lab, Drug Mechanism Enrichment Analysis (DMEA), can successfully identify effective drug combinations for cancer cells. Leveraging public databases of transcriptomic profiles (e.g., Cancer Cell Line Encyclopedia) and drug sensitivity scores (e.g., Cancer Dependency Map), we analyzed the sensitivity of >300 cancer cell lines to >1,300 drugs to define the global landscape of drug sensitivity. First, we calculated Pearson correlation estimates between the sensitivity scores (AUC) for each drug. Next, we ran DMEA with these Pearson correlation estimates to identify drug mechanisms (e.g., EGFR inhibitors, RAF inhibitors, etc.) enriched based on sensitivity to each drug. Lastly, we ran DMEA again, now ranking the drugs based on the NES scores for each drug mechanism. Finally, the resulting NES scores identified drug mechanisms which may benefit cells sensitive or resistant to each of 85 drug mechanisms. This approach validated that our method can successfully predict sensitivity to ninety nine percent of annotated drug mechanisms. Next, to identify potential drug combinations from our global analysis, we compared the strength and direction of each pair of drug mechanisms. The pairings with the greatest positive scores represented the top potential drug combinations for sensitive cancer cell lines. In our analysis, the top drug combination was inhibitors of polo-like kinase (PLK) and tubulin polymerization. In support of this finding, published studies indicate that tubulin inhibition can increase the toxicity of PLK inhibitors in various cancers. This suggests that other drug combinations predicted by our algorithm may also exhibit synergy against cancer cells. Taken together, our study presents a wealth of information about potential drug combinations for cancer and represents a framework for future analysis of large drug screen datasets.